Radiative effects of anthropogenic changes in composition of the atmosphere are expected to intensify global water cycling. Translation to increased flood risk is speculative and has been questioned on the basis of the observational record; recent climate-change assessments have limited their hydrologic scope mainly to precipitation characteristics. Here we address the flood problem using a climate model, with a river-discharge submodel, that is shown to simulate successfully the occurrence of extratropical great floods (discharges in excess of the 100-y event from river basins of area greater than 200,000 km²). For the 14 basins considered, a quadrupling of atmospheric CO₂ increases the probability of occurrence of great floods by a factor having a median value greater than 5. For the same set of basins, observed great-flood frequency increased during the 20th century. The significance of the observed increase is evaluated through an analysis of an ensemble of 5 transient climate-change experiments and a 900-y control experiment. The analysis suggests that the observed increase could be explained by unforced variability; that the forced increase in the model during the same period is relatively small; and that clear detection of an increase in great-flood frequency could be 20-60 years away.